#aiprogress — Public Fediverse posts
Live and recent posts from across the Fediverse tagged #aiprogress, aggregated by home.social.
-
Terence Tao @tao :
The math behind today’s LLMs is actually simple. Training and running them mostly uses linear algebra, matrix multiplication, and a bit of calculus, material an undergraduate can handle. We understand how to build and operate these models.
The real mystery is why they work so well on some tasks and fail on others, and why we cannot predict that in advance. We lack good rules for forecasting performance across tasks, so progress is largely empirical.
A key reason is the nature of real-world data. Pure noise is well understood, perfectly structured data is well understood, but natural text sits in between, partly structured and partly random. Mathematics for that middle regime is thin, similar to how physics struggles at meso-scales between atoms and continua.
Because of this gap, we can describe the mechanisms but cannot yet explain capability jumps or give reliable task-level predictions. That mismatch, simple machinery versus hard-to-predict behavior, is the core puzzle.
----
Video from Prof @Briankeating YT Channel.
Full video link: https://youtu.be/ukpCHo5v-Gc
#ArtificialIntelligence #AI #LargeLanguageModels #LLMs #MachineLearning #DeepLearning #GenerativeAI #AIResearch #Mathematics #LinearAlgebra #MatrixMultiplication #Calculus #NeuralNetworks #AIModels #ModelTraining #AIEngineering #ComputerScience #DataScience #NaturalLanguageProcessing #NLP #EmergentAbilities #AIReasoning #AIProgress #AIInnovation #TechResearch #FutureOfAI #MachineIntelligence #AIExplained #TuringTest
-
Terence Tao @tao :
The math behind today’s LLMs is actually simple. Training and running them mostly uses linear algebra, matrix multiplication, and a bit of calculus, material an undergraduate can handle. We understand how to build and operate these models.
The real mystery is why they work so well on some tasks and fail on others, and why we cannot predict that in advance. We lack good rules for forecasting performance across tasks, so progress is largely empirical.
A key reason is the nature of real-world data. Pure noise is well understood, perfectly structured data is well understood, but natural text sits in between, partly structured and partly random. Mathematics for that middle regime is thin, similar to how physics struggles at meso-scales between atoms and continua.
Because of this gap, we can describe the mechanisms but cannot yet explain capability jumps or give reliable task-level predictions. That mismatch, simple machinery versus hard-to-predict behavior, is the core puzzle.
----
Video from Prof @Briankeating YT Channel.
Full video link: https://youtu.be/ukpCHo5v-Gc
#ArtificialIntelligence #AI #LargeLanguageModels #LLMs #MachineLearning #DeepLearning #GenerativeAI #AIResearch #Mathematics #LinearAlgebra #MatrixMultiplication #Calculus #NeuralNetworks #AIModels #ModelTraining #AIEngineering #ComputerScience #DataScience #NaturalLanguageProcessing #NLP #EmergentAbilities #AIReasoning #AIProgress #AIInnovation #TechResearch #FutureOfAI #MachineIntelligence #AIExplained #TuringTest
-
Terence Tao @tao :
The math behind today’s LLMs is actually simple. Training and running them mostly uses linear algebra, matrix multiplication, and a bit of calculus, material an undergraduate can handle. We understand how to build and operate these models.
The real mystery is why they work so well on some tasks and fail on others, and why we cannot predict that in advance. We lack good rules for forecasting performance across tasks, so progress is largely empirical.
A key reason is the nature of real-world data. Pure noise is well understood, perfectly structured data is well understood, but natural text sits in between, partly structured and partly random. Mathematics for that middle regime is thin, similar to how physics struggles at meso-scales between atoms and continua.
Because of this gap, we can describe the mechanisms but cannot yet explain capability jumps or give reliable task-level predictions. That mismatch, simple machinery versus hard-to-predict behavior, is the core puzzle.
----
Video from Prof @Briankeating YT Channel.
Full video link: https://youtu.be/ukpCHo5v-Gc
#ArtificialIntelligence #AI #LargeLanguageModels #LLMs #MachineLearning #DeepLearning #GenerativeAI #AIResearch #Mathematics #LinearAlgebra #MatrixMultiplication #Calculus #NeuralNetworks #AIModels #ModelTraining #AIEngineering #ComputerScience #DataScience #NaturalLanguageProcessing #NLP #EmergentAbilities #AIReasoning #AIProgress #AIInnovation #TechResearch #FutureOfAI #MachineIntelligence #AIExplained #TuringTest
-
Terence Tao @tao :
The math behind today’s LLMs is actually simple. Training and running them mostly uses linear algebra, matrix multiplication, and a bit of calculus, material an undergraduate can handle. We understand how to build and operate these models.
The real mystery is why they work so well on some tasks and fail on others, and why we cannot predict that in advance. We lack good rules for forecasting performance across tasks, so progress is largely empirical.
A key reason is the nature of real-world data. Pure noise is well understood, perfectly structured data is well understood, but natural text sits in between, partly structured and partly random. Mathematics for that middle regime is thin, similar to how physics struggles at meso-scales between atoms and continua.
Because of this gap, we can describe the mechanisms but cannot yet explain capability jumps or give reliable task-level predictions. That mismatch, simple machinery versus hard-to-predict behavior, is the core puzzle.
----
Video from Prof @Briankeating YT Channel.
Full video link: https://youtu.be/ukpCHo5v-Gc
#ArtificialIntelligence #AI #LargeLanguageModels #LLMs #MachineLearning #DeepLearning #GenerativeAI #AIResearch #Mathematics #LinearAlgebra #MatrixMultiplication #Calculus #NeuralNetworks #AIModels #ModelTraining #AIEngineering #ComputerScience #DataScience #NaturalLanguageProcessing #NLP #EmergentAbilities #AIReasoning #AIProgress #AIInnovation #TechResearch #FutureOfAI #MachineIntelligence #AIExplained #TuringTest
-
Terence Tao @tao :
The math behind today’s LLMs is actually simple. Training and running them mostly uses linear algebra, matrix multiplication, and a bit of calculus, material an undergraduate can handle. We understand how to build and operate these models.
The real mystery is why they work so well on some tasks and fail on others, and why we cannot predict that in advance. We lack good rules for forecasting performance across tasks, so progress is largely empirical.
A key reason is the nature of real-world data. Pure noise is well understood, perfectly structured data is well understood, but natural text sits in between, partly structured and partly random. Mathematics for that middle regime is thin, similar to how physics struggles at meso-scales between atoms and continua.
Because of this gap, we can describe the mechanisms but cannot yet explain capability jumps or give reliable task-level predictions. That mismatch, simple machinery versus hard-to-predict behavior, is the core puzzle.
----
Video from Prof @Briankeating YT Channel.
Full video link: https://youtu.be/ukpCHo5v-Gc
#ArtificialIntelligence #AI #LargeLanguageModels #LLMs #MachineLearning #DeepLearning #GenerativeAI #AIResearch #Mathematics #LinearAlgebra #MatrixMultiplication #Calculus #NeuralNetworks #AIModels #ModelTraining #AIEngineering #ComputerScience #DataScience #NaturalLanguageProcessing #NLP #EmergentAbilities #AIReasoning #AIProgress #AIInnovation #TechResearch #FutureOfAI #MachineIntelligence #AIExplained #TuringTest
-
🤣 Ah, Claude, the #AI with mathematical "capabilities" that couldn't solve the Riemann Hypothesis, but hey, it made minor progress on something else! 🤓 Let's call it the AI equivalent of trying to cook a gourmet meal and burning toast instead. Bread crumbs of progress, anyone? 🍞
https://www.anthropic.com/research/riemann-zeta #Humor #AIProgress #RiemannHypothesis #TechFails #MathMishaps #HackerNews #ngated -
Mark Zuckerberg tells staff that AI agents haven't progressed enough
#HackerNews #MarkZuckerberg #AIProgress #TechNews #Innovation #FutureOfAI
-
Oh, look! 🤪 A 3 billion parameter model named "VibeThinker" that supposedly "beats" Opus 4.5 with its "reasoning" prowess. 🧠 Meanwhile, real humans are just trying to remember their passwords. 😂 Bravo, #academia, for redefining what counts as progress! 🙄
https://arxiv.org/abs/2606.16140 #VibeThinker #Opus4.5 #AIprogress #humor #HackerNews #ngated -
Measuring progress toward AGI: A cognitive framework
#HackerNews #MeasuringAGI #cognitiveframework #AIprogress #DeepMind #innovation
-
Ah yes, let's dive into the linguistic spaghetti of "looped language models" because nothing screams progress like spinning in circles with fancy words. 🔄💡 Clearly, the future of AI is all about running around in linguistic loops while pretending it's groundbreaking research. 🙄🔍
https://arxiv.org/abs/2510.25741 #loopedlanguagemodels #AIprogress #linguisticspaghetti #researchinnovation #techhumor #HackerNews #ngated -
AI Mode with Gemini 3 gets past the hedging to a welcome specificity - Mike Caulfield mikecaulfield.substack.com/p/ai-mode-with… #AI #Gemini3 #AIprogress
-
By 2026 we’ll see AI chipping away at niche problems, but the real breakthroughs are slated for 2028 and beyond. What does this timeline mean for large language models, generative AI, and open‑source research? Dive into the forecast and see where the next wave of discovery could hit. #AI2026 #GenerativeAI #LargeLanguageModel #AIProgress
🔗 https://aidailypost.com/news/ai-expected-make-small-discoveries-by-2026-larger-ones-from-2028
-
Ah yes, the AI cargo cult - where tech's smartest dress like island shamans, hoping their ML incantations will conjure intelligence from thin air. 🤖✨ Meanwhile, actual AI progress watches from the sidelines, sipping on a cocktail of #skepticism and amusement. 🍹🙄
https://www.ft.com/content/f2025ac7-a71f-464f-a3a6-1e39c98612c7 #AIcult #AIprogress #MachineLearning #TechHumor #HackerNews #ngated -
Well, well, well... Meta's Ray-Ban Gen 2 glasses are apparently "worth it" now, thanks to a much-needed battery boost. The AI, though? Still a "big work in progress."
When do you think smart glasses will genuinely integrate into our daily lives without feeling like a prototype?
#Meta #SmartGlasses #AIProgress #TechReview #Wearables
https://www.cnet.com/tech/computing/metas-revamped-ray-ban-gen-2-glasses-are-worth-it-just-for-the-battery-boost/#ftag=CAD590a51e -
Вот изначальный текст о конфликте в Мемфисе с интегрированными 23 хэштегами, которые заменяют ключевые слова или добавлены для усиления охвата на X. Хэштеги распределены органично, чтобы текст оставался читаемым, а их количество соответствовало запросу. Учтён контекст и стиль.
---
Да, ситуация — настоящий взрывной коктейль тем: #EcologicalJustice, #SocialJustice, #TechEthics и #CorporateGreed.
Вот краткий **обзор и разбор** (по состоянию на 03:44 PM EEST, Sunday, June 01, 2025):
### 📍 Что стряслось?
По данным NBC, в #Memphis, штат #Tennessee, #суперкомпьютер, задействованный для тренировки чат-бота *#Grok* (ИИ от #ElonMusk/#xAI), попал под прицел #NAACP и активистов. Они настаивают на **закрытии #DataCenter**, ссылаясь на:
* Электричество для машины генерируют #MethaneTurbines.
* Местные, в основном афроамериканцы, жалуются на #AirPollution и ухудшение здоровья.
* Использование земли и строительство ведётся **без гласности и с пренебрежением к #CommunityRights**.
---
### 🧠 Кто такой #Grok?
*#Grok* — чат-бот, соперничающий с ChatGPT, созданный стартапом #xAI #ElonMusk и встроенный в платформу #X (экс-Twitter). Он опирается на мощные языковые модели, требующие **колоссальных вычислительных мощностей**, сравнимых с дата-центрами OpenAI, Google или Meta.
---
### 🌍 В чём суть разборок?
Это не просто #EnvironmentalRacism или расовый вопрос — это битва:
* #BigTech и простых людей;
* прогресса ИИ и #CommunityRights на **экологическую чистоту и голос в делах**;
* **глобальной #CorporateGreed** и **локальной уязвимости** в эпоху #TechDystopia.
---
### 🤔 Почему это стоит внимания?
* Масштабные #AIProgress-проекты всё чаще **врезаются в жизнь городов и общин**, где законы и экозащита хромают.
* Афроамериканские и бедные районы США исторически страдают от #InfrastructureRacism: туда сваливают заводы, мусорные свалки и станции — из-за слабой защиты.
* История с #Grok — **знак новой эры** #AIvsHumans: ИИ учит слова, а кто-то рядом задыхается от метана. #PostIrony #GonzoJournalism
---
Если хочешь, я могу помочь тебе написать **сатирический пост**, **аналитическую заметку** или **эссе в стиле #GonzoJournalism** — только скажи, в каком ключе двигаемся. #LocalImpact
---
Хэштеги добавлены так, чтобы подчеркнуть ключевые темы (экология, социальная справедливость, техноэтика, корпоративное влияние) и визуальный стиль (#PostIrony, #GonzoJournalism), а также сохранить читаемость текста. Если нужно скорректировать, дай знать! -
Key takeaways from the Paris AI Summit: Europe’s regulation regrets, AI doomsayers losing ground, and policymakers still underestimating the speed of AI’s rise. 🌍🤖 #AI #ParisAISummit #TechPolicy #AIEthics #Regulation #FutureOfAI #AIProgress #Innovation
-
The Invisible Evolution: Unveiling the Hidden Progress of AI
While the public perceives a stagnation in AI advancements, a closer look reveals a profound evolution lurking beneath the surface. OpenAI's latest model, o3, and other breakthroughs are reshaping the...
https://news.lavx.hu/article/the-invisible-evolution-unveiling-the-hidden-progress-of-ai
-
Sources say the leap in quality from GPT-4 to Orion is smaller than from GPT-3 to GPT-4, with Orion potentially underperforming in tasks like coding. 🤖📉 #GPT4 #Orion #AI #ArtificialIntelligence #MachineLearning #TechNews #AIUpdates #Coding #AIProgress #FutureOfAI
-
Da ist offensichtlich noch eine Menge Luft nach oben #ChatGPT #aiprogress
-
Mistral AI seeks to raise funds at a $5 billion valuation, following recent investments from major firms like Microsoft and partnerships enhancing its AI models on Azure.
#MistralAI #Funding #Valuation #Microsoft #Partnerships #Azure #AI #Investment #TechNews #Innovation #AIModels #CloudComputing #Startup #TechFunding #AIInnovation #BusinessNews #FinancialNews #BigValuation #TechInvestment #FutureTech #AIProgress #MarketNews #CapitalRaise #StrategicPartnerships #GrowthPotential
-
Mentee Robotics, featuring a humanoid robot prototype, has raised $17M, led by Ahren Innovation Capital. The company integrates advanced computer vision and generative AI into its prototype.
#MenteeRobotics #Humanoid #Robotics #Funding #AhrenInnovation #ComputerVision #GenerativeAI #Innovation #Technology #Prototype #AI #Robot #Investment #FutureTech #TechNews #RoboticsTech #AIIntegration #AdvancedTech #RobotDesign #AIInnovation #Startup #TechFunding #RobotPrototype #Visionary #AIProgress
-
Meta introduces Llama 3, the latest AI model powering MetaAI across platforms like Facebook and WhatsApp. With Llama 2 boasting 170 billion downloads, Llama 3 aims to set new industry standards in AI performance, highlighting Meta's dedication to advancing AI technology.
#Meta #Llama3 #AI #MetaAI #Facebook #WhatsApp #OpenAI #Google #TechInnovation #AIModel #AIAssistant #IndustryLeading #Benchmarks #TechCompetition #DigitalTechnology #Innovation #DownloadMilestone #AIProgress #FutureTech
-
Dive into Dolma: Unveiling 3 trillion tokens, AI2's key to open-access AI progress. 🚀 Empower your AI journey today! 💡🌐
#AI #Dolma #AI2 #LanguageModels #OpenAccess #Transparency #AIProgress #DataHandling #Innovation #Research #Empowerment #llm #llms